Despite commonalities in core social communication traits, young children on the autism spectrum show significant phenotypic heterogeneity and variability in response to interventions focused on improving social communication. There is a considerable dearth of psychometrically sound, sensitive and objective measures of early social communication behaviors that can be implemented across the range of heterogeneity associated with the autism phenotype.

Commonly implemented standardized assessments vary in sensitivity to subtle behavioral changes that may occur after a short period of intervention. In contrast, non-standardized observational measures that have revealed meaningful treatment effects tend to be based on proximal social communication behaviors directly targeted during the intervention. Because behaviors are treatment-specific, effects may not be easily generalizable to broader or more distal, developmentally downstream outcomes such as later language and social communication development. Most critically, the rigorous manual annotation process needed for most proximal outcomes is expensive and time-intensive, challenging the feasibility of translating such observational measures into scalable protocols for widespread community implementation. As autism science accelerates into the digital era, there is a prime opportunity to develop accessible mobile technology to improve the objectivity, granularity and scalability of remote automated measurement of core social communication behaviors in young children on the autism spectrum.

In the current project, Rachel Reetzke, Rebecca Landa and colleagues plan to harness advances in computer vision which have enabled automatic human movement tracking (i.e., “pose estimation”) from videos recorded using common household devices (e.g., smartphones, tablets) with minimal cost, time investment and technological requirements. This project leverages the well-phenotyped SPARK (Simons Foundation Powering Autism Research) cohort and a retrospective longitudinal home video approach to (1) refine and evaluate the validity, test-retest reliability and feasibility of a movement-based Automated Measure of Early Social Communication (AMES) and (2) examine the predictive validity of AMES versus standardized measures of early social communication at 36 months of age for distal language and social communication outcomes at school-age.

The novel data collected from this study has the potential to advance (a) accurate and rapid early autism phenotyping, (b) fine-grained intervention progress monitoring and uniform comparison of treatment efficacy across autism early interventions and (c) large-scale analyses of genotype-phenotype associations.